Open Source

The tools below are built in the open. Each embodies practices described on this site: specs as code, zero-dependency substrates, agent-ready workflows.

Drift

github.com/SteelSprint/Drift

• A sync layer between your specs and your code: the agent writes both, Drift keeps them aligned.
• Specs are plain-English rules in XML files; markers are comment lines that wrap the implementing code.
• Language-agnostic; works with any text file.
• Single static binary, zero dependencies.
• drift skill teaches an LLM agent the whole workflow: write specs, place markers, link, check, resolve.
• Dogfoods itself; its own build fails if any spec drifts from its code.
• This site is built with it.

luv

github.com/SteelSprint/luv

• "The conversation is the data structure": a canonical conversation type for LLM applications.
• Structure-preserving transformations ("morphisms") carry one conversation value to and from real provider APIs; no per-provider adapter code at the call site.
• Switch providers mid-conversation: OpenAI, Anthropic, and AWS Bedrock Converse morphisms shipped.
• Conversations are trees (id + parent_id): forks, regenerations, and agent branches are first-class.
• A substrate, not a framework: bring your own orchestration, memory, retries, and cost tracking.
• Streaming and non-streaming reduce to the same Reply type; errors are data.
• Zero third-party dependencies; TypeScript reference implementation on web-standard APIs.